Figure 3.

Probability density estimate of the true error for LDA classification rule, without feature selection (left column, d =10 global markers are directly used), with t -test feature selection (right column), ρ =0 . 8, COV =0 . 05,$\sum _j{X}_{i,j}\in (50\mathrm{K},100\mathrm{K})$ for SAGE-tags and the following total number of reads for NGS-reads: (a,b)$\sum _j{X}_{i,j}\in (1\mathrm{K},50\mathrm{K})$; (c,d)$\sum _j{X}_{i,j}\in (250\mathrm{K},300\mathrm{K})$; (e,f)$\sum _j{X}_{i,j}\in (25\mathrm{M},25\mathrm{M}+50\mathrm{K})$; (g,h)$\sum _j{X}_{i,j}\in (50\mathrm{M},50\mathrm{M}+50\mathrm{K})$.